[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[PULL 2/7] i386/xen: fix per-vCPU upcall vector for Xen emulation
From: |
David Woodhouse |
Subject: |
[PULL 2/7] i386/xen: fix per-vCPU upcall vector for Xen emulation |
Date: |
Mon, 6 Nov 2023 10:39:50 +0000 |
From: David Woodhouse <dwmw@amazon.co.uk>
The per-vCPU upcall vector support had three problems. Firstly it was
using the wrong hypercall argument and would always return -EFAULT when
the guest tried to set it up. Secondly it was using the wrong ioctl() to
pass the vector to the kernel and thus the *kernel* would always return
-EINVAL. Finally, even when delivering the event directly from userspace
with an MSI, it put the destination CPU ID into the wrong bits of the
MSI address.
Linux doesn't (yet) use this mode so it went without decent testing
for a while.
Cc: qemu-stable@nongnu.org
Fixes: 105b47fdf2d0 ("i386/xen: implement HVMOP_set_evtchn_upcall_vector")
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
---
target/i386/kvm/xen-emu.c | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/target/i386/kvm/xen-emu.c b/target/i386/kvm/xen-emu.c
index 0055441b2e..7c504d9fa4 100644
--- a/target/i386/kvm/xen-emu.c
+++ b/target/i386/kvm/xen-emu.c
@@ -306,7 +306,7 @@ static int kvm_xen_set_vcpu_callback_vector(CPUState *cs)
trace_kvm_xen_set_vcpu_callback(cs->cpu_index, vector);
- return kvm_vcpu_ioctl(cs, KVM_XEN_HVM_SET_ATTR, &xva);
+ return kvm_vcpu_ioctl(cs, KVM_XEN_VCPU_SET_ATTR, &xva);
}
static void do_set_vcpu_callback_vector(CPUState *cs, run_on_cpu_data data)
@@ -440,7 +440,8 @@ void kvm_xen_inject_vcpu_callback_vector(uint32_t vcpu_id,
int type)
* deliver it as an MSI.
*/
MSIMessage msg = {
- .address = APIC_DEFAULT_ADDRESS | X86_CPU(cs)->apic_id,
+ .address = APIC_DEFAULT_ADDRESS |
+ (X86_CPU(cs)->apic_id << MSI_ADDR_DEST_ID_SHIFT),
.data = vector | (1UL << MSI_DATA_LEVEL_SHIFT),
};
kvm_irqchip_send_msi(kvm_state, msg);
@@ -849,8 +850,7 @@ static bool kvm_xen_hcall_hvm_op(struct kvm_xen_exit *exit,
X86CPU *cpu,
int ret = -ENOSYS;
switch (cmd) {
case HVMOP_set_evtchn_upcall_vector:
- ret = kvm_xen_hcall_evtchn_upcall_vector(exit, cpu,
- exit->u.hcall.params[0]);
+ ret = kvm_xen_hcall_evtchn_upcall_vector(exit, cpu, arg);
break;
case HVMOP_pagetable_dying:
--
2.41.0
- [PULL 0/7] xenfv-stable queue, David Woodhouse, 2023/11/06
- [PULL 3/7] hw/xen: select kernel mode for per-vCPU event channel upcall vector, David Woodhouse, 2023/11/06
- [PULL 2/7] i386/xen: fix per-vCPU upcall vector for Xen emulation,
David Woodhouse <=
- [PULL 4/7] hw/xen: don't clear map_track[] in xen_gnttab_reset(), David Woodhouse, 2023/11/06
- [PULL 1/7] i386/xen: Don't advertise XENFEAT_supervisor_mode_kernel, David Woodhouse, 2023/11/06
- [PULL 7/7] hw/xen: use correct default protocol for xen-block on x86, David Woodhouse, 2023/11/06
- [PULL 6/7] hw/xen: take iothread mutex in xen_evtchn_reset_op(), David Woodhouse, 2023/11/06
- [PULL 5/7] hw/xen: fix XenStore watch delivery to guest, David Woodhouse, 2023/11/06
- Re: [PULL 0/7] xenfv-stable queue, Stefan Hajnoczi, 2023/11/06